ali-vilab / VGen

Official repo for VGen: a holistic video generation ecosystem for video generation building on diffusion models
https://i2vgen-xl.github.io
2.75k stars 243 forks source link

Re-training model from checkpoints #113

Open nothonfz opened 2 months ago

nothonfz commented 2 months ago

I find there is a PRETRAIN.build() func after the initialization of Unet in line 154 of train_t2v_enterance.py. I think this code is used to load SD parameters and I did not find the code for re-training model from checkpoint. Was it my oversight that I didn't find the relevant code?